80,000 Hours Podcast cover image

Risks from power-seeking AI systems (article narration by Zershaaneh Qureshi)

80,000 Hours Podcast

00:00

Why evaluations may miss deception

Zershaaneh Qureshi describes limits of interpretability and testing, including fake alignment, sandbagging, hidden reasoning, and sleeper-agent behavior.

Play episode from 39:36
Transcript

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app